Gear shaft machining defect intelligent detection system based on deep learning

Abstract

The invention relates to a gear shaft processing defect intelligent detection system based on deep learning, which belongs to the technical field of intelligent manufacturing and quality control, and comprises a multi-modal sensing data acquisition module used for acquiring multi-modal time sequence signals in a processing process to generate an original sensing data stream; the process-morphology cross-scale mapping module is used for receiving an original sensing data stream and mapping the original sensing data stream into a deep causal state feature sequence; and the defect generation prediction and traceability module is used for combining the deep causal state feature sequence and the instantaneous space coordinates of the cutter to calculate the conditional probability of a specific type of defects, and constructing a defect probability space distribution diagram covering the machining surface of the workpiece. The fundamental problem that in the prior art, only real-time process data can be analyzed, but the specific shape and position of the gear shaft cannot be accurately predicted before defects are formed is solved, and the machining yield of the high-precision gear shaft is greatly improved.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of intelligent manufacturing and quality control, and particularly relates to a gear shaft machining defect intelligent detection system based on deep learning. BACKGROUND

[0002] In the automatic machining of high-precision gear shafts, the control of product surface quality is a core link. Traditional quality control methods are usually divided into two independent categories: process monitoring and post-detection. The process monitoring system judges whether the machining state is abnormal by analyzing the statistical characteristics of sensor signals such as cutting force and vibration, but cannot predict what specific surface defects will be caused by the abnormality. Post-detection is to identify and classify the formed defects after machining by using optical or non-destructive detection equipment. At this time, the defective products have already been produced, causing waste of materials and working hours. The prior art fails to establish a deterministic causal relationship between dynamic sensing data in the machining process and the micro-defect morphology finally formed, so it cannot effectively warn and intervene before the defects are formed. SUMMARY

[0004] The application aims to provide a gear shaft machining defect intelligent detection system based on deep learning to solve the problems in the background art.

[0005] The technical scheme of the application comprises a multi-modal sensing data acquisition module for acquiring multi-modal time series signals in the machining process to generate an original sensing data stream;

[0006] A process-morphology cross-scale mapping module is used to receive the original sensing data stream and map it into a deep causal state feature sequence;

[0007] A defect generative prediction and tracing module is used to combine the deep causal state feature sequence and the instantaneous spatial coordinates of the tool to calculate the conditional probability of a specific type of defect and construct a defect probability spatial distribution map covering the machining surface of the workpiece;

[0008] A machining process self-adaptive regulation and control module is used to compare the defect probability in the defect probability spatial distribution map with a preset risk threshold. When the defect probability exceeds the risk threshold, a regulation and control instruction is generated and executed. When the defect probability does not exceed the risk threshold, the current machining process is maintained.

[0009] Preferably, the multi-modal time series signals comprise cutting force signals, three-axis vibration acceleration signals and acoustic emission signals.

[0010] Preferably, the specific processing steps of the process-morphology cross-scale mapping module are as follows:

[0011] The instantaneous spectral entropy of each type of signal in the original sensing data stream is calculated;

[0012] calculating a normalized cross-correlation function value between different physical signals;

[0013] combining the preset entropy weight coefficient and the cross-correlation weight coefficient, and performing weighted summation on the instantaneous spectrum entropy and the normalized cross-correlation function value to generate a deep causal state feature sequence.

[0014] Preferably, the entropy weight coefficient and the cross-correlation weight coefficient are determined by principal component analysis or sensitivity analysis on historical data.

[0015] Preferably, the calculation step of the conditional probability is:

[0016] based on the deep causal state feature sequence under the historical processing state, performing exponential decay weighting and integration along the time dimension to calculate a weighted cumulative damage value of the historical state;

[0017] inputting the weighted cumulative damage value into a Sigmoid activation function for mapping to generate the conditional probability.

[0018] Preferably, the sensitivity rate weight, the forgetting factor and the bias term parameter used in the calculation step are obtained by training a deep learning model on a labeled data set containing process data and three-dimensional surface topography scanning data.

[0019] Preferably, before generating the control instruction, the processing technology self-adaptive control module is further used for:

[0020] analyzing the time corresponding to the deep causal state feature that contributes most in the defect probability rising stage;

[0021] extracting the original sensing data corresponding to the time for spectral characteristic analysis to generate defect trace information identifying the defect cause.

[0022] Preferably, the control instruction is an instruction for fine-tuning the spindle speed to avoid the resonance region according to the specific frequency vibration identified in the defect trace information.

[0023] Preferably, the risk threshold is determined by analyzing the receiver operating characteristic curve of the historical data after balancing the cost of defective products and the loss of production efficiency.

[0024] The present application provides a gear shaft processing defect intelligent detection system based on deep learning, which has the following advantages:

[0025] The application constructs a complete closed-loop intelligent control system, including a multi-modal sensing data acquisition module, a process-morphology cross-scale mapping module, a defect generative prediction and tracing module and a processing technology self-adaptive regulation and control module; the system can monitor the processing process in real time, accurately predict the type and spatial position of the workpiece surface defects, and actively regulate the process based on the defect causes, so as to eliminate the defects before they are formed; this solves the fundamental problem that the prior art can only analyze real-time process data, but cannot accurately predict the specific morphology and position of the defects before they are formed, greatly improving the processing yield of high-precision gear shafts;

[0026] The multi-modal sensing data acquisition module of the application obtains physical information capable of comprehensively representing the dynamic interactive state of the tool-workpiece-machine tool system by synchronously collecting cutting force signals, three-axis vibration acceleration signals and acoustic emission signals; compared with single signal monitoring, the multi-modal data combination with complementary physical connotation provides more abundant and robust original basis for subsequent feature extraction, significantly enhancing the sensing ability of the system to complex processing states and the reliability of defect prediction;

[0027] The process-morphology cross-scale mapping module of the application can extract a low-dimensional and high-information-density deep causal state feature sequence from high-dimensional and noisy original sensing data stream; the module objectively and accurately reveals the internal risk of the processing system tending to form surface microscopic defects by structurally quantifying the instantaneous spectral entropy of each signal and the normalized cross-correlation function between different signals, and combining the entropy weight coefficient and the cross-correlation weight coefficient determined by principal component analysis or sensitivity analysis for weighted summation; this data-driven feature generation method avoids the subjectivity of manually setting parameters, so that the feature can more effectively represent the deterioration trend of the processing state;

[0028] The defect generative prediction and tracing module of the application associates the deep causal state feature sequence in the time dimension with the tool instantaneous spatial coordinates by establishing a probability model simulating the physical process of damage accumulation; the module exponentially decays and weights the historical state and integrates along the time dimension, calculates the weighted cumulative damage value of the historical state, and maps it to the conditional probability of a specific type of defect; this method not only accurately simulates the physical nature of defect formation, but also can construct a defect probability spatial distribution map covering the entire workpiece processing surface; the key sensitivity rate weight, forgetting factor and bias term parameters in the model are obtained by deep learning model training, ensuring the high accuracy and strong adaptability of the prediction result;

[0029] The processing technology self-adaptive regulation and control module realizes precise and efficient closed-loop intervention; before generating the regulation and control instruction, the module can analyze the moment corresponding to the depth causal state feature that contributes most in the defect probability rising stage, and perform spectrum characteristic analysis on the original sensing data at the moment to generate defect traceability information identifying the defect cause; accordingly, the system can generate a targeted regulation and control instruction, for example, fine-tune the spindle speed according to the identified specific frequency vibration to avoid the resonance zone; such precise regulation and control based on traceability realizes treatment according to the disease, efficiently solves the problem with the smallest process adjustment amplitude, and maximally reduces the negative impact on production efficiency.

[0030] The risk threshold for decision-making in the application is scientifically determined by analyzing the receiver operating characteristic curve of historical data, balancing the cost of defective products and the loss of production efficiency; this systematic setting method overcomes the subjectivity and non-optimality of traditional experience setting threshold, so that the decision-making behavior of the system can reach the best balance point according to the economic requirements of production, thereby significantly improving the overall production economic benefits. BRIEF DESCRIPTION OF DRAWINGS

[0032] The application will be further explained below in combination with the drawings and embodiments:

[0033] Figure 1 is a flow chart of a gear shaft machining defect intelligent detection system based on deep learning. DETAILED DESCRIPTION

[0035] In order to make the purpose, technical scheme and advantages of the application clearer, the application will be further described in detail below in combination with specific embodiments.

[0036] Embodiment 1:

[0037] Please refer to Figure 1 The application provides a gear shaft machining defect intelligent detection system based on deep learning, comprising:

[0038] The multi-modal sensing data acquisition module is used for acquiring multi-modal time series signals in the machining process to generate an original sensing data stream;

[0039] The process-topography cross-scale mapping module is used for receiving the original sensing data stream and mapping it into a depth causal state feature sequence;

[0040] The defect generative prediction and traceability module is used for combining the depth causal state feature sequence and the instantaneous spatial coordinates of the tool to calculate the conditional probability of a specific type of defect, and to construct a defect probability spatial distribution map covering the machining surface of the workpiece;

[0041] The processing process self-adaptive regulation and control module is used for comparing the defect probability in the defect probability space distribution map with a preset risk threshold value, generating and executing a regulation and control instruction when the defect probability exceeds the risk threshold value, and maintaining the current processing process when the defect probability does not exceed the risk threshold value.

[0042] The multi-modal sensing data acquisition module is deployed on the gear shaft machining equipment; the purpose of the module is to capture physical signals that can comprehensively reflect the dynamic interaction state of the tool-workpiece-machine tool system in the machining process in real time and synchronously; in the embodiment, the module acquires multi-modal time series signals at a high frequency by installing various sensors at key positions of the machine tool, and the signals are integrated to form an original sensing data stream and are transmitted to a subsequent processing module in real time;

[0043] The process-topography cross-scale mapping module receives the original sensing data stream; the purpose of the module is to extract a low-dimensional but higher information density feature from high-dimensional and noisy original sensing data, and the feature can directly represent the inherent risk of the machining system to form surface micro-defects; in the embodiment, the module realizes this purpose by calculating a deep causal state feature The deep causal state feature is a single time series scalar that quantifies the complexity and coupling of the system under multiple physical fields, and its role is to realize data dimension reduction and to make the nonlinear mode such as resonance or instability directly related to defect formation explicit; the calculation formula of the feature is defined as:

[0044]

[0045] The definitions of the parameters in the formula are as follows:

[0046] is the sum of all sensor signal types;

[0047] is the sum of all different types of sensor signal pairs;

[0048] is the deep causal state feature output by the module at time is a dimensionless scalar, and the numerical fluctuation represents the risk degree of the machining system to form defects;

[0049] is the type index of the sensor signal, for example, representing cutting force, X-axis vibration, etc.

[0050] is the i-th The entropy weight coefficient of the signal type is a dimensionless scalar. This parameter is determined by principal component analysis or sensitivity analysis of historical data and reflects the contribution of the complexity of different types of signals to the formation of defects.

[0051] For the first in the original sensor data stream The signal calculated at time 10:00 The instantaneous spectral entropy is a dimensionless value. This value is obtained by calculating the information entropy of the normalized power spectral density within a certain time window after performing a short-time Fourier transform on the signal, and it reflects the frequency complexity of the signal at the current moment.

[0052] For signal and The cross-correlation weighting coefficient between them is a dimensionless scalar; the source of this parameter is related to... Similarly, historical data analysis is used to determine the contribution of the coupling effect between specific signal pairs to defect formation;

[0053] For the signal in the original sensing data stream and The calculated time delay The normalized cross-correlation function value is a dimensionless value used to quantify the coupling strength and response delay between two different physical signals;

[0054] The time delay parameter is determined by analyzing the sequence of different signal responses in historical data to capture the causal lag effect in the physical process.

[0055] Time delay parameter It is determined by calculating the peak value of the cross-correlation function between different signal pairs at different delays, and its value is taken as the delay that makes the cross-correlation function reach its maximum value;

[0056] Calculated The time series data is used as direct input to the subsequent prediction module;

[0057] The defect generation prediction and tracing module receives the deep causal state feature sequence generated by the previous module. It is the time from the start of processing. up to the current moment The module aims to establish a deterministic mapping relationship between the state evolution in the time dimension of the processing process and the defects ultimately formed in the spatial dimension of the workpiece, thereby achieving online defect prediction. In this embodiment, the module obtains the state of the CNC system of the processing equipment in real time, thus achieving online defect prediction. Instantaneous spatial coordinates of the tool acting on the workpiece surface ; it utilizes an integral probability model with a forgetting factor to associate temporal features with instantaneous spatial coordinates , which essentially simulates the cumulative damage effect in physics; this model is used to calculate the conditional probability of a specific type of defect occurring at the current machining position , which is mathematically expressed as follows:

[0058]

[0059] The definitions of the parameters in the formula are as follows:

[0060] is the conditional probability, which is a dimensionless value representing the possibility of the occurrence of the first type of defect at the current tool position given the historical machining state (where represents the morphology, represents the category, represents the position vector);

[0061] : Sigmoid activation function, which maps the internal cumulative damage value to the probability interval of ;

[0062] is the sensitivity rate weight for the first type of defect, with the physical dimension of time inverse ; this parameter is obtained through model training, and it quantifies the contribution intensity of the causal state feature in unit time to the logarithmic occurrence rate of a specific defect type;

[0063] is the forgetting factor, also known as the decay rate, with the physical dimension of time inverse ; this parameter is also obtained through model training, and it determines the speed of historical state information decay, reflecting the memory length of the machining system to past disturbances;

[0064] is the exponential decay kernel function, which is dimensionless, making recent states have higher weights than distant states;

[0065] is the deep causal state feature at a certain time in the past calculated by the pre-process - morphology cross-scale mapping module;

[0066] ​​is a bias term, is a dimensionless learnable parameter, represents the inherent tendency of the i-th feature to cause the process to deviate from the nominal state without any process disturbance. Inherent tendency of the i-th feature to cause the process to deviate from the nominal state without any process disturbance.

[0067] As the machining progresses, the module will associate each calculated probability value with its corresponding spatial coordinates, and ultimately construct a complete high-resolution defect probability spatial distribution map covering the machined surface of the workpiece. Spatial coordinates corresponding thereto

[0068] The machining process adaptive control module receives the defect probability spatial distribution map output by the prediction module; the purpose of this module is to actively and intelligently adjust the machining process parameters before the actual formation of defects according to the predicted defect risk, in order to avoid or mitigate the occurrence of defects in a closed-loop manner; in this embodiment, the module will compare the real-time generated defect probability with the preset risk threshold; [risk threshold] refers to a decision boundary for determining whether the defect occurrence probability reaches the degree of intervention; when the defect probability exceeds the risk threshold, the module will automatically generate and execute control instructions; when the defect probability does not exceed the risk threshold, the current machining process is maintained to avoid unnecessary intervention affecting production efficiency.

[0069] The system disclosed in this embodiment solves the fundamental problem that existing technologies cannot accurately predict the type, size, and location of defects before their formation by analyzing real-time process data, by constructing a complete mapping and prediction link from multi-modal sensor data to deep causal state features, and then to defect spatial probability distribution, and combining closed-loop adaptive control; it changes traditional post-detection to pre-prediction and intervention, and can eliminate defect hazards before defective products are produced, thereby greatly improving the yield of high-precision gear shaft machining, reducing material and labor waste, and achieving truly intelligent quality control.

[0070] Embodiment 2:

[0071] The multi-modal time series signals include cutting force signals, three-axis vibration acceleration signals, and acoustic emission signals.

[0072] On the basis of the system of embodiment 1, the specific collection content of the multi-modal sensor data collection module is limited; in this embodiment, the multi-modal time series signals are specifically at least three types of key physical signals: cutting force signals, three-axis vibration acceleration signals, and acoustic emission signals.

[0073] The cutting force signal is collected by a force sensor installed on the tool holder or spindle, and reflects the interaction force between the tool and the workpiece during the cutting process; its function is to directly represent the cutting load and tool state; ​​

[0074] Triaxial vibration acceleration signals are X, Y, Z three-direction vibration signals collected by accelerometers installed on key components of machine tools, such as spindle boxes and worktables, and the role is to comprehensively monitor the dynamic stability of the machine tool system;

[0075] Acoustic emission signals are transient elastic wave signals released by the generation or expansion of micro cracks in the material, which are collected by acoustic sensors installed on the workpiece or fixture near the cutting area, and the role is to have extremely high sensitivity to micro damage events in the material;

[0076] The technical effect of gain is that by simultaneously collecting three signals with different sources and complementary physical connotations, the state information of the processing system can be more comprehensively and robustly captured; the cutting force directly reflects the processing load, the vibration signal reflects the system stability, and the acoustic emission signal can reveal early damage at the material level; compared with single signal monitoring, this multi-modal combination can provide more abundant and more distinguishable original information for subsequent deep causal state feature extraction, thereby improving the accuracy and reliability of defect prediction.

[0077] Embodiment 3:

[0078] The specific processing steps of the process-topography cross-scale mapping module are:

[0079] Calculate the instantaneous spectral entropy of each type of signal in the original sensing data stream;

[0080] Calculate the normalized cross-correlation function value between different physical signals;

[0081] Combine the preset entropy weight coefficient and cross-correlation weight coefficient to weight and sum the instantaneous spectral entropy and the normalized cross-correlation function value to generate a deep causal state feature sequence; the present application fuses the instantaneous frequency complexity of each signal and the coupling strength between signals by weighting, which integrates the dynamic interaction information of multiple physical fields in the system, thereby being able to represent the evolution trend of the processing state from a higher dimension; the depth of the feature lies in that it reveals the nonlinear mapping relationship between the original signal and the final defect through a data-driven weight determination method; the causality lies in that it can capture the dynamic lag effect in the processing process;

[0082] The entropy weight coefficient and the cross-correlation weight coefficient are determined by principal component analysis or sensitivity analysis on historical data;

[0083] On the basis of the system of embodiment 1, the specific calculation steps of the process-topography cross-scale mapping module for generating a deep causal state feature sequence are described in detail; the specific processing process of the module is as follows: first, process each type of signal in the original sensing data stream to calculate its instantaneous spectral entropy ; the inner logic is to quantify the frequency complexity or uncertainty of each type of signal at the current time, the more chaotic and disordered the signal spectrum, the higher the entropy value, which is usually related to the instability of the processing state; second, the normalized cross-correlation function value between different physical signals is calculated , which aims to quantify the coupling strength and response timing relationship between different physical phenomena; based on the above calculation results, the instantaneous spectral entropy and the cross-correlation function value are multiplied by the corresponding weight coefficients and respectively and summed up, and finally fused into a comprehensive state indicator, i.e. deep causal state feature ;

[0084] To ensure the objectivity and effectiveness of the fusion process, the entropy weight coefficient and the cross-correlation weight coefficient are determined by principal component analysis or sensitivity analysis on historical data; principal component analysis or sensitivity analysis refers to a data-driven parameter optimization method, which works as follows: on a historical data set containing a large amount of processing process data and final product surface quality labels, the correlation strength between the spectral entropy of each signal and the cross-correlation between signals and the final defect formation is analyzed, and the optimal weight coefficient combination is learned;

[0085] The technical effect of the gain is that this step-by-step calculation and weighted fusion method not only structurally extracts deep information from two dimensions of signal complexity and signal coupling degree, but also determines the weight coefficient in a data-driven way, avoiding the subjectivity and blindness of manual parameter setting; this makes the finally generated deep causal state feature can more objectively and accurately reflect the deterioration trend of the processing state, laying a solid data foundation for subsequent accurate prediction and improving the intelligent level of the entire system and the generalization ability of the model.

[0086] Embodiment 4:

[0087] The calculation steps of the conditional probability are as follows:

[0088] Based on the deep causal state feature sequence under the historical processing state, the exponential decay weighting is performed and integrated along the time dimension to calculate the weighted cumulative damage value of the historical state;

[0089] The weighted cumulative damage value is input into the Sigmoid activation function for mapping to generate the conditional probability;

[0090] The sensitivity rate weight, forgetting factor and bias term parameters used in the calculation steps are obtained by deep learning model training on a labeled data set containing process data and three-dimensional surface topography scanning data;

[0091] The specific steps of the defect generative prediction and tracing module calculating conditional probability are described in detail on the basis of the embodiment 1 system; the calculation process first performs exponential decay weighting based on the deep causal state feature sequence under the historical processing state and integrates along the time dimension; the internal mechanism is that for each historical time in the sequence The feature value is weighted using an exponential decay kernel function , so that the closer the state, the greater the current impact; then, the weighted feature value is multiplied by the sensitivity rate weight and integrated along the time axis; the integration result is added to a bias term together to form the weighted cumulative damage value of the historical state; this step aims to simulate the physical process of defect formation, that is, the final macroscopic defect is the result of continuous accumulation of historical microscopic damage; then, the weighted cumulative damage value calculated in the previous step is input into the Sigmoid activation function for mapping, which converts an internal, physically meaningful damage degree into a standard, conditional probability in the interval;

[0092] For further clarification, the three core parameters used in the calculation step, namely the sensitivity rate weight , the forgetting factor and the bias term , are obtained by deep learning model training on a labeled data set containing process data and three-dimensional surface topography scanning data; deep learning model training refers to a process in which a large amount of historical processing sensing data is used as input, and the real type and location of the workpiece surface defect corresponding to the data are used as labels, which are obtained by high-precision three-dimensional scanners, through optimization algorithms such as back propagation, automatically adjusting model parameters to minimize the difference between the defect probability distribution predicted by the model and the real defect distribution;

[0093] The integral probability model is implemented through a recurrent neural network or a long short-term memory network; the network takes the deep causal state feature sequence as input, and its internal state is responsible for simulating the cumulative damage value, and finally outputs the conditional probability through the Sigmoid activation function; the in the formula is the weight, decay rate and bias term parameter of the network, which is obtained by end-to-end training on the labeled data set;

[0094] The deep learning model in this invention is a multilayer perceptron or recurrent neural network, trained using the Adam optimizer and binary cross-entropy loss function; the training dataset is a labeled dataset containing historical processing sensing data and corresponding three-dimensional surface topography scanning data; the training objective is to minimize the difference between the defect probability distribution predicted by the model and the actual defect distribution.

[0095] The gain technique enables this integral-based damage accumulation model to accurately simulate the physical nature of defect formation, making the prediction results more interpretable. More importantly, by obtaining model parameters through end-to-end deep learning training, the model can automatically learn from the data the sensitivity of different types of defects to process disturbances, the system's memory effect, and the inherent occurrence rate, which greatly improves the accuracy of the prediction model and its adaptability to different processing conditions, achieving highly customized and automated defect prediction.

[0096] Example 5:

[0097] Before generating control commands, the adaptive control module for machining processes is also used for:

[0098] Analyze the moment when the deep causal state feature contributes the most to the increase in defect probability.

[0099] The original sensor data corresponding to the time point is extracted and its spectral characteristics are analyzed to generate defect source information that identifies the cause of the defect.

[0100] The control command is generated based on the specific frequency vibration identified in the defect tracing information, and is used to fine-tune the spindle speed to avoid the resonance zone.

[0101] The risk threshold is determined by analyzing the receiver's operational characteristic curves of historical data, after balancing the cost of defective products with the loss of production efficiency.

[0102] Based on the system in Example 1, the defect tracing function of the adaptive control module for processing technology before generating control commands is described in detail; when the module detects the probability of a defect at a certain location... When a risk begins to rise significantly and shows a tendency to exceed the risk threshold, a source analysis will be performed before generating regulatory instructions. This analysis process includes back-calculating the integral of the probability value and identifying the historical moments that made the major contribution to the integral term. These moments are considered critical points that lead to an increased risk of current defects; subsequently, the module uses the identified critical moments... The system precisely retrieves signal segments before and after a given moment from the stored raw data stream and performs detailed spectral analysis on them to identify any abnormal frequency components. The analysis results are then integrated into defect tracing information that identifies the cause of the defect.

[0103] In a specific application scenario of this embodiment, the control command is generated based on the specific frequency vibration identified in the defect tracing information. It is used to fine-tune the spindle speed to avoid the resonance zone. For example, if the tracing information shows that the main reason for the increase in the defect probability is the surge in vibration energy at a specific frequency, and this frequency is close to a certain harmonic of the current spindle speed, the system can determine that cutting resonance has occurred. Accordingly, the control module will automatically generate a command and send it to the CNC system of the machine tool. The command content is to fine-tune the spindle speed, thereby changing the excitation frequency of the system and actively destroying the resonance condition.

[0104] Based on the system in Example 1, the risk threshold in the adaptive control module for processing technology is adjusted. The setting method is described in detail; the risk threshold is described in detail. It is not a fixed value set based on experience, but rather the optimal decision point determined by analyzing the receiver's operational characteristic curve of historical data and comprehensively considering the defective product costs caused by underreporting and the production efficiency losses caused by false reporting.

[0105] ROC curve analysis is a statistical method for evaluating the performance of binary classifiers. In this embodiment, its application logic is as follows: First, on a dataset containing a large number of historical processed samples, by iterating through all possible probability values ​​as thresholds, the true positive rate (the proportion of correctly predicted defects) and false positive rate (the proportion of incorrectly classified defects) are calculated at each threshold. Then, an ROC curve is plotted with the false positive rate on the horizontal axis and the true positive rate on the vertical axis. Finally, a cost-benefit analysis is introduced to quantify the costs of missed and false positives, and combined with the true positive and false positive rates at each point on the ROC curve, the total expected cost at each threshold is calculated. The point that minimizes the total expected cost is selected, and its corresponding probability value is determined as the optimal risk threshold. ;

[0106] The gain technology achieves its effect by determining risk thresholds in a systematic and data-driven manner, overcoming the subjectivity and non-optimal nature of traditional methods that rely on human experience to set thresholds. This method can scientifically find an optimal balance point based on the company's specific trade-offs between quality costs and production efficiency. This optimizes the system's decision-making behavior, ensuring that while effectively preventing the generation of defective products, unnecessary interference with normal production is minimized, thereby improving the overall economic efficiency of production.

[0107] It should be noted that the above examples are only used to illustrate the technical solutions of the present application but not limit the present application. Although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalently replaced, without departing from the spirit and scope of the technical solutions of the present application.

Claims

1. A deep learning-based gear shaft machining defect intelligent detection system, characterized in that, The method comprises the following steps: a multi-modal sensor data acquisition module is used to collect multi-modal time series signals in the machining process to generate raw sensor data streams; a process-topography cross-scale mapping module is used to receive the raw sensor data streams and map them into deep causal state feature sequences; a defect generative prediction and tracing module is used to combine the deep causal state feature sequences with the instantaneous spatial coordinates of the tool to calculate the conditional probability of a specific type of defect and construct a defect probability spatial distribution map covering the machined surface of the workpiece; a machining process self-adaptive regulation and control module is used to compare the defect probability in the defect probability spatial distribution map with a preset risk threshold value, and when the defect probability exceeds the risk threshold value, a regulation and control instruction is generated and executed, and when the defect probability does not exceed the risk threshold value, the current machining process is maintained.

2. The deep learning-based gear shaft machining defect intelligent detection system according to claim 1, wherein The multi-modal time series signals include cutting force signals, three-axis vibration acceleration signals, and acoustic emission signals.

3. The deep learning-based gear shaft machining defect intelligent detection system according to claim 1, characterized in that, The specific processing steps of the process-topography cross-scale mapping module are as follows: the instantaneous spectral entropy of each type of signal in the raw sensor data stream is calculated; the normalized cross-correlation function value between different physical signals is calculated; the instantaneous spectral entropy and the normalized cross-correlation function value are weighted and summed by combining the preset entropy weight coefficient and the cross-correlation weight coefficient to generate the deep causal state feature sequence.

4. The deep learning-based gear shaft machining defect intelligent detection system according to claim 3, characterized in that, The entropy weight coefficient and the cross-correlation weight coefficient are determined by principal component analysis or sensitivity analysis on historical data.

5. The deep learning-based gear shaft machining defect intelligent detection system according to claim 1, characterized in that, The calculation steps of the conditional probability are as follows: based on the deep causal state feature sequence under the historical machining state, the weighted cumulative damage value of the historical state is calculated by exponential decay weighting and integration along the time dimension; the weighted cumulative damage value is input into the Sigmoid activation function for mapping to generate the conditional probability.

6. The deep learning-based gear shaft machining defect intelligent detection system according to claim 5, characterized in that, The sensitivity rate weight, the forgetting factor, and the bias term parameters used in the calculation steps are obtained by training a deep learning model on a labeled data set containing process data and three-dimensional surface topography scanning data.

7. The deep learning-based gear shaft machining defect intelligent detection system according to claim 1, characterized in that, Before generating the regulation and control instruction, the machining process self-adaptive regulation and control module is also used to: analyze the time corresponding to the deep causal state feature that contributes most to the rising stage of the defect probability; extract the original sensor data corresponding to the time for spectral characteristic analysis to generate defect tracing information that identifies the causes of the defect.

8. The deep learning-based gear shaft machining defect intelligent detection system according to claim 7, characterized in that, The regulation and control instruction is an instruction for fine-tuning the spindle speed to avoid the resonance region according to the specific frequency vibration identified in the defect tracing information.

9. The deep learning-based gear shaft machining defect intelligent detection system according to claim 1, wherein The risk threshold value is determined by analyzing the receiver operating characteristic curve of the historical data after balancing the cost of defective products and the loss of production efficiency.

Images